阅读说明：本技术 一种气体泄漏源强反算改进型搜索算法 (Improved search algorithm for strong back calculation of gas leakage source ) 是由 武海丽 关磊 魏利军 朱敬聪 赵倩琳 朱晓光 于 2020-01-20 设计创作，主要内容包括：本发明公开了一种气体泄漏源强反算改进型搜索算法,包括以下步骤：使用传感器网络获取泄漏现场的浓度测量数据,先验扩散模型计算泄漏现场的浓度计算数据,通过建立测量数据和先验扩散模型间的函数关系,将反算问题转化为最优化求解问题,利用改进型遗传算法完成泄漏源参数预估值；将改进型遗传算法的预估值结果作为马尔科夫链蒙特卡洛抽样初始抽样点,利用马尔科夫链蒙特卡洛抽样算法得到泄漏源位置的概率分布范围；以马尔科夫链蒙特卡洛抽样算法提供的泄漏源位置的分布范围,使用移动机器人进场主动嗅探进行处理,利用改进型保守收敛粒子群算法用于完成泄漏源精确定位。先利用基于传感器网络的定位技术确定疑似的泄漏源区域,而后在疑似区域内有针对性的释放移动机器人进行进一步搜索,取长补短联合处理危害气体泄漏事故中源强反算问题。(The invention discloses an improved search algorithm for strong back calculation of a gas leakage source, which comprises the following steps: the method comprises the steps of obtaining concentration measurement data of a leakage site by using a sensor network, calculating concentration calculation data of the leakage site by using a prior diffusion model, converting a back calculation problem into an optimized solution problem by establishing a functional relation between the measurement data and the prior diffusion model, and completing leakage source parameter pre-estimation by using an improved genetic algorithm; taking the pre-estimated value result of the improved genetic algorithm as a Markov chain Monte Carlo sampling initial sampling point, and obtaining the probability distribution range of the position of the leakage source by utilizing the Markov chain Monte Carlo sampling algorithm; the distribution range of the positions of the leakage sources provided by the Markov chain Monte Carlo sampling algorithm is used, the mobile robot is used for entering the field and actively sniffing for processing, and the improved conservative convergence particle swarm algorithm is used for finishing accurate positioning of the leakage sources. Firstly, a suspected leakage source area is determined by utilizing a positioning technology based on a sensor network, then, the mobile robot is released in the suspected area in a targeted manner for further searching, and the source strength back calculation problem in the accident of harmful gas leakage is jointly processed by taking the advantages and the disadvantages.)

1. An improved search algorithm for strong back calculation of a gas leakage source is characterized by comprising the following steps:

the method comprises the steps of obtaining concentration measurement data of a leakage site by using a sensor network, calculating concentration calculation data of the leakage site by using a prior diffusion model, converting a back calculation problem into an optimized solution problem by establishing a functional relation between the measurement data and the prior diffusion model, and completing leakage source parameter pre-estimation value calculation by applying an improved genetic algorithm;

taking the pre-estimated value result of the improved genetic algorithm as an initial sampling point of a Markov chain Monte Carlo sampling algorithm, and obtaining the probability distribution range of the position of the leakage source by utilizing the Markov chain Monte Carlo sampling algorithm;

the distribution range of the positions of the leakage sources provided by the Markov chain Monte Carlo sampling algorithm is used, the leakage sources are searched in the distribution range of the leakage sources by using a mobile robot, and the accurate positioning of the leakage sources is completed by using the improved conservative convergence particle swarm algorithm.

2. The improved search algorithm for gas leakage source strong back calculation according to claim 1, wherein the obtaining of the concentration measurement data of the leakage site by using the sensor network, the calculation of the concentration calculation data of the leakage site by using the prior diffusion model, the transformation of the back calculation problem into the optimized solution problem by establishing the functional relationship between the measurement data and the prior diffusion model, and the calculation of the leakage source parameter estimated value by using the improved genetic algorithm specifically comprise:

the measured value of the sensor network to the ith measuring position isCalculating the calculated concentration of the ith measurement position by using a gas diffusion modelSolving an objective function using an improved genetic algorithmIs measured.

3. The gas leak source strength back-calculation improved search algorithm as claimed in claim 2, wherein the calculated concentration at the ith measurement position is calculated using a gas diffusion modelThe method specifically comprises the following steps:

calculating the calculated concentration of the ith measurement position by adopting a gas diffusion model Passell-Gifford Gaussian modelFor surface instantaneous leakage, a model of the mass of smoke from a point source on the surface is used, i.e.

In the formula (I), the compound is shown in the specification,

x_i,y_i,t_iis the coordinates of the ith measurement location,for given positions (x, y) andleakage concentration in kg/m at time t³；

Q is the mass of the instantaneously leaked materials, and the unit is kg;

u is wind speed;

t is leakage time;

the diffusion coefficients in x, y and z directions, respectively, are given in m.

4. The gas leak source strength back-calculation improved search algorithm of claim 2, wherein the applying an improved genetic algorithm to solve the objective functionSpecifically, the minimum value of (2) includes:

collecting the unit sample individuals of the leakage source discarded in the selection step in each iteration, and storing the discarded unit sample individuals of the leakage source into an eliminator gene library for expanding the crossing range;

and collecting unit sample individuals of each iteratively optimal leakage source, enhancing the evolution probability of the population towards the optimal point, and simultaneously carrying out active local search operation by taking the current optimal point as a starting point when the update of the optimal point is stagnated until the number of newly generated leakage object particles meets the requirement.

5. The improved search algorithm for gas leakage source intensity back-calculation as claimed in claim 4, wherein the obtaining of the probability distribution range of the leakage source position by using the markov chain monte carlo sampling algorithm specifically comprises:

and (3) using an adaptive Metropolis algorithm as a sampling algorithm of Markov chain Monte Carlo sampling, and performing data processing on the candidate points by adopting logarithmic balance until a sampling termination condition is met.

6. The improved search algorithm for the strong back calculation of the gas leakage source according to claim 5, wherein the improved conservative convergent particle swarm algorithm is used for achieving accurate positioning of the leakage source, and specifically comprises:

the method comprises the steps that operation is carried out according to a standard particle swarm algorithm at the initial stage until a historical optimal value of a particle swarm updates a stagnation algebra to exceed a set limit value;

calling partial particles and historical optimal value particles from subgroups with lower fitness in the same generation to form an optimal group, wherein the optimal group is used for enhancing local search strength near an optimal value point;

the position and the speed of the particles with smaller fitness are reset away from the optimal point, so that the algorithm is guaranteed to jump out of a local extreme value as far as possible;

and dynamically updating the optimal clique until the optimal clique successfully contracts and positions or an iterative algebra reaches an upper limit.

Technical Field

The invention relates to the technical field of dangerous chemical gas leakage detection, in particular to a strong back calculation improved search algorithm for a gas leakage source.

Background

The determination of the position and the intensity of the dangerous chemical gas leakage source is a cornerstone for emergency rescue of dangerous chemical gas leakage accidents, and the research in the field is of importance and urgency. Hazardous chemical gas leaks are often sudden, resulting in uncertainty as to the time of the leak, the location of the leak, and the environment of the leak. In emergency situations, the unknown leakage source is required to be determined in the shortest time under the limited condition of information loss, so as to further determine the emergency evacuation area and the safety distance and provide a basis for emergency decision. After dangerous chemical gas leaks, monitoring equipment is utilized to monitor gas concentration on the leakage site and around the leakage site, and then the position of the leakage source is located by utilizing monitored data. The leakage source positioning mode is mainly divided into two types: and monitoring, calculating and positioning by a sensor network and searching and positioning by a mobile robot in a coordinated manner.

The idea of monitoring, calculating and positioning the sensor network is to integrate a prior gas diffusion model and gas concentration data acquired in real time to solve the gas source parameter iterative solution distributed estimation problem. The method is realized by firstly obtaining simulated concentration according to a candidate parameter set through a prior model, establishing a matching function of the simulated concentration and the measured concentration of the sensor network through a Bayes theory, and continuously optimizing the result of the matching function to obtain the posterior distribution of the gas source parameters.

The idea of mobile robot collaborative search positioning is that a mobile robot is used for installing a corresponding gas concentration sensor to form an individual with olfaction perception, so that the mobile robot search which is due to the problem of strong inverse calculation is also called as an active olfaction problem, and the robot uses wind direction information and gas concentration information and adopts an heuristic search strategy to realize positioning search. The leakage source positioning based on the sensor network is limited by data limitation of the sensor network and fixity of network arrangement, and the leakage source positioning based on cooperation of the mobile robot can fully exert motion flexibility of the mobile robot to realize high-detail and more-targeted local area search, so that the scene adaptability of the mobile robot is stronger, and the mobile robot can cope with a more complex and dynamic environment.

The following research shortcomings exist for the gas source monitoring and positioning technology based on the wireless sensor network:

(1) the data processing algorithm belonging to the statistical probability distribution category needs to be based on accident information and prior information of parameters to be solved, although the uncertainty of the inversion result can be effectively evaluated, the sampling process of posterior distribution is time-consuming, and the selection of the initial sampling point is directly related to the accuracy and the non-accuracy of the sampling result;

(2) the inverse algorithm belonging to the optimal solution category has strong dependence on the prior diffusion model, and due to the high nonlinearity of the source intensity inverse calculation problem, the inverse calculation can be performed only by adopting the direct optimization algorithm, but the direct optimization algorithm mainly based on the genetic algorithm needs a very large population scale to obtain a relatively ideal result, which directly results in the increase of the calculation cost index;

(3) the balance problem between the scale of the sensor network and the accuracy of the searched target is as follows: obviously, to obtain the suspected source parameters with higher accuracy, the scale of the sensor network must be increased, which is a typical balance between cost and effort.

The gas source monitoring and positioning technology based on the mobile robot has the following technical defects:

(1) dynamic distribution characteristics of the smoke plume caused by turbulence. The small-scale vortex tears the smell/gas smoke plume into a plurality of filaments, so that the concentration distribution numerical value of a local micro area is oscillated, and a local extreme value trap is easily brought to the mobile robot; the crushing phenomenon of the actual gas diffusion field is obvious, so that the mobile robot has lower adaptability to a search space with larger scale;

(2) the algorithm efficiency of the current multi-robot system is generally low, the efficient heuristic strategy is absent, and few researches relate to parallel positioning under the multi-source scene, so that the scene adaptability of the mobile robot for collaborative searching is greatly limited.

Disclosure of Invention

The invention aims to provide an improved search algorithm for gas leakage source intensity back calculation, which is characterized in that a suspected leakage source area is determined by utilizing a positioning technology based on a sensor network, then a mobile robot is released in the suspected area in a targeted manner for further search, and the source intensity back calculation problem in the accident of damaging gas leakage is treated by taking the best of the best and the best.

In order to achieve the purpose, the invention adopts the following technical scheme:

an improved search algorithm for gas leakage source strong back calculation comprises the following steps:

acquiring concentration measurement data of a leakage site by using a sensor network, calculating concentration calculation data of the leakage site by using a prior diffusion model, converting a back calculation problem into an optimized solution problem by establishing a functional relation between the measurement data and the prior diffusion model, and completing leakage source parameter pre-estimation by using an improved genetic algorithm;

taking the pre-estimated value result of the improved genetic algorithm as a Markov chain Monte Carlo sampling initial sampling point, and obtaining the probability distribution range of the position of the leakage source by utilizing the Markov chain Monte Carlo sampling algorithm;

the distribution range of the positions of the leakage sources provided by the Markov chain Monte Carlo sampling algorithm is used, the mobile robot is used for entering the field and actively sniffing for processing, and the improved conservative convergence particle swarm algorithm is used for finishing the accurate positioning of the leakage sources.

Further, the obtaining of the concentration measurement data of the leakage site by using the sensor network, the calculating of the concentration data of the leakage site by using the prior diffusion model, and the transforming of the back calculation problem into the optimized solution problem by establishing a functional relationship between the measurement data and the prior diffusion model specifically include:

the measured value of the sensor network to the ith measuring position isCalculating the calculated concentration of the ith measurement position by using a gas diffusion modelSolving an objective functionIs measured.

Further, the implementation of the leakage source parameter estimation using the improved genetic algorithm specifically includes:

collecting the unit sample individuals of the leakage source discarded in the selection step in each iteration, and storing the discarded unit sample individuals of the leakage source into a knockout gene library independently for expanding the crossing range;

and collecting unit sample individuals of the leakage source with optimal iteration in each step, introducing a 'following and searching' strategy, enhancing the evolution probability of the population towards the optimal point, and simultaneously carrying out active local searching operation by taking the current optimal point as a starting point when the optimal point is updated to be stagnated until the number of newly generated leakage source particles meets the requirement.

Further, the obtaining of the probability distribution range of the position of the leakage source by using the markov chain monte carlo sampling algorithm specifically includes:

the self-adaptive Metropolis algorithm is used as a sampling algorithm of Markov chain Monte Carlo sampling, and a 'logarithmic balance' data processing and 'partial matching and collaborative searching' strategy is adopted until a sampling termination condition is met.

Further, the accurate positioning of the leakage source is completed by using an improved conservative convergence particle swarm algorithm, and specifically includes:

the method comprises the steps that operation is carried out according to a standard particle swarm algorithm at the initial stage until a historical optimal value of a particle swarm updates a stagnation algebra to exceed a set limit value;

calling partial particles and historical optimal value particles from subgroups with lower fitness in the same generation to form an optimal group, wherein the optimal group is used for enhancing local search strength near an optimal value point;

the position and the speed of the particles with smaller fitness are reset away from the optimal point, so that the algorithm is guaranteed to jump out of a local extreme value as far as possible;

and dynamically updating the optimal clique until the optimal clique successfully contracts and positions or an iterative algebra reaches an upper limit.

Drawings

FIG. 1 is a flow chart of an improved search algorithm for gas leak source strength back calculation according to the present invention;

FIG. 2 is a new cross-flow diagram of the improved genetic algorithm MGA of the present invention;

FIG. 3 is one embodiment of step S2 of FIG. 2;

fig. 4 is one embodiment of step S3 in fig. 2.

Detailed Description

In order to clearly explain the technical features of the present invention, the following detailed description of the present invention is provided with reference to the accompanying drawings. The following disclosure provides many different embodiments, or examples, for implementing different features of the invention. To simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Furthermore, the present invention may repeat reference numerals and/or letters in the various examples. This repetition is for the purpose of simplicity and clarity and does not in itself dictate a relationship between the various embodiments and/or configurations discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and procedures are omitted so as to not unnecessarily limit the invention.

As shown in fig. 1, an improved search algorithm for gas leakage source strength back calculation includes the following steps:

s1, acquiring concentration measurement data of a leakage site by using a sensor network, calculating concentration calculation data of the leakage site by using a prior diffusion model, converting a back calculation problem into an optimized solution problem by establishing a functional relation between the measurement data and the prior diffusion model, and completing leakage source parameter estimation values by using an improved genetic algorithm;

s2, taking the pre-estimated value result of the improved genetic algorithm as a Markov chain Monte Carlo sampling initial sampling point, and obtaining the probability distribution range of the position of the leakage source by utilizing the Markov chain Monte Carlo sampling algorithm;

s3, processing the leakage source position distribution range provided by the Markov chain Monte Carlo sampling algorithm by using mobile robot approach active sniffing, and using the improved conservative convergence particle swarm algorithm to finish accurate positioning of the leakage source.

As an embodiment of the present application, in step S1, obtaining concentration measurement data of a leakage site by using a sensor network, calculating concentration calculation data of the leakage site by using a prior diffusion model, and transforming a back-calculation problem into an optimization solution problem by establishing a functional relationship between the measurement data and the prior diffusion model, specifically including:

the measured value of the sensor network to the ith measuring position isCalculating the calculated concentration of the ith measurement position by using a gas diffusion modelSolving an objective functionIs measured.

Wherein the content of the first and second substances,calculated by adopting a gas diffusion model Passell-Gifford Gaussian model, and for the ground instantaneous leakage, a tobacco mass model of a ground instantaneous point source is adopted, namely

In the formula (I), the compound is shown in the specification,

x_i,y_i,t_iis the coordinates of the ith measurement location,leakage concentration in kg/m for a given location (x, y) and time t³；

Q is the mass of the instantaneously leaked material, and the unit is kg;

u is wind speed;

t is leakage time;

the diffusion coefficients in x, y and z directions, respectively, are given in m.

As shown in fig. 2, as an embodiment of the present application, a new cross flow of the modified genetic algorithm MGA is as follows:

(1) randomly selecting two different individuals from the subgroup reserved in the 'selection' stage, and respectively marking as P₁,P₂I.e. selecting the unit sample of leakage that remains.

(2) Generating a random number r₁U (0,1), let us note that the crossover rate is α, if r₁< α, then P₁,P₂Performing cross operation according to the conventional cross flow of the GA; if r is₁If not less than α, turning to the step (3);

(3) recording the current optimal point update stagnation algebra of the system as S, setting the upper limit of the stagnation algebra of the optimal point as SetMax, if S is less than SetMax, entering the step (4), otherwise, entering the step (5);

(4) randomly selecting two individuals from the 'selection' rejected gene bank, and respectively marking as P_s1,P_s2I.e. selecting rejected units of leakage for matching P separately₁,P₂Performing a crossover operation in which P₁,P₂Crossing as a parent, and recording the inheritance rate of the parent as β;

(5) generating a random number r₁U (0,1), the tracking rate is gamma, if r₁If the gamma value is less than gamma, entering the step (6), otherwise, entering the step (7);

(6) the following process: note that the current optimal particle is P_best,P₁,P₂Is operated as followsLine follow, where f is a random number and f-N (0, 1):

P_{new_i}＝P_best+f×(P_best-P_i)

(7) active searching: note that the current optimal particle is P_best,P₁,P₂The active search is performed as follows:

P_{new_i}＝β×P_i+P_best×(1-β)

(8) and (4) repeating the steps (1) to (7) until the number of newly generated leakage particles is required.

Compared with the traditional GA cross flow design, the main improvement of MGA is as follows: collecting the unit sample individuals of the leakage source discarded in the selection step in each iteration, and storing the discarded unit sample individuals of the leakage source into a knockout gene library for expanding the crossing range;

and collecting unit sample individuals of the leakage source with optimal iteration in each step, introducing a 'following and searching' strategy, enhancing the evolution probability of the population towards the optimal point, and simultaneously carrying out active local searching operation by taking the current optimal point as a starting point when the update of the optimal point is stagnated until the number of newly generated leakage particles meets the requirement.

As shown in fig. 3, as an embodiment of the present application, in step S2, the obtaining a probability distribution range of the leakage source position by using the markov chain monte carlo sampling algorithm specifically includes:

s21, initializing Markov chain with the sampling starting point of theta₀＝(x₀,y₀,Q₀) Recording the target parameter point obtained in the t step as theta_t＝(x_t,y_t,Q_t) Wherein X is_t,Y_tAnd are the coordinates of the measurement location.

The MCMC method expresses model errors, measurement errors, prior distribution of parameters and the like in a probability density function mode, and more reasonably describes the uncertainty of a source strength back calculation problem through the distribution of sampling results. When the search space of the target object is complex, the MCMC is sensitive to the selection of the initial point and the proposed distribution, and an unreasonable initial point may cause a large number of invalid samples of the mahalanobis chain outside the convergence domain. The result of the pre-estimation using MGA is therefore the initial starting point of sampling of the MCMC.

S22, at the current point theta_t＝(x_t,y_t,Q_t) As mean center, covariance matrix C_tObtaining a random candidate point theta^ca～N(θ_t,C_t)，C_tCan be expressed as:

s23, according toCalculates the acceptance rate α of the candidate parameter points and generates a random number rand-U (0,1) with the acceptance rule of

And S24, repeating S22 and S23 until the sampling termination condition is met, if the sampling algebra meets the requirement.

As shown in fig. 4, as an embodiment of the present application, in step S3, the method uses a mobile robot approach active sniffing to process a distribution range of leakage source positions provided by a markov chain monte carlo sampling algorithm, and uses an improved conservative convergent particle swarm algorithm to complete accurate positioning of a leakage source, which specifically includes:

s31, initializing parameters of the particle swarm, namely the total number of particles, the coordinate point and the initial velocity of the particles of the leakage sample;

s32, operating according to a standard particle swarm algorithm at the initial stage until the historical optimal value of the particle swarm updates the stagnation algebra to exceed the set limit;

s33, the algorithm enters a 'calling and resetting' stage: noting that the particles with historical optimum value areDistributing M auxiliary particles from the particles with smaller fitness, wherein the M +1 particles become an optimal group;

particles of S34 "optimum clusters" according toCarrying out speed updating; part of the particles are selected from the particles with smaller adaptability in the present generation as the reset speed and the position of the free particles, and the rest of the particles still follow the speed of the standard PSOUpdating is carried out;

in the formula (I), the compound is shown in the specification,

V_i: the flight speed of the particles;

P_i: the best location that the individual particles have experienced;

P_g: the best position for all particles in the whole population to pass through;

w: an inertia factor, the value of which is non-negative; the value is large, the global optimizing capability is strong, and the local optimizing capability is weak; the value is large, the global optimizing capability is weak, and the local optimizing capability is strong; therefore, the larger w is, the larger the speed and the position updating amplitude of the particle flying are, and the larger the deviation degree of the original optimized orbit is;

ρ: a range parameter;

C₁，C₂: acceleration constant, C1 is a constant judged from the experience of the individual, C2 is the experience of the population; both are learning factors, usually C₁＝C₂＝2；

r₁,r₂: is [0,1 ]]Random numbers transformed within a range;

s35, updating the new position of the particle swarm according to the updated speed, then calculating the new adaptive value of the particle swarm, updating the self optimal historical value of the particle swarm, then updating the optimal global historical value of the particle swarm, and if the new optimal global value appears in the common sub-swarm, directly and completely setting the current existing 'optimal group' sub-swarm as an auxiliary particle swarm of new optimal particles, which is the 'optimal group' dynamic updating operation;

s36, if the common subgroup has no new global optimum value, updating the search result record of the optimal group according to whether the search of the new optimum value of the optimal group is successful, and updating the local search range parameter rho according to whether the search result record meets the continuous success or continuous failure times;

s37, repeating the steps S34, S35 and S36 until the optimal clique successfully shrinks the positioning or the iterative algebra reaches the upper limit.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.